Retrotransposition Events Shape the Evolution of the Ataxin-3 Gene Family in Primates

Abstract

Evolutionary studies of disease-associated genes provide crucial insights into pathological mechanisms and potential therapeutic targets. Polyglutamine spinocerebellar ataxias (SCAs) are human neurodegenerative diseases caused by toxic expanded CAG repeats. Studies on SCA1 have shown that a paralog of the causing-gene can partially rescue protein function and alleviate the neuropathology. The most common SCA, Machado–Joseph disease (MJD/SCA3), caused by mutated ataxin-3 gene (ATXN3), has no treatment currently available. Its paralog ataxin-3 like (ATXN3L) remains largely unexplored. Here, we identify three new retrotransposition events of ATXN3: ATXN3L0 in Euarchontoglires, ATXN3L2 in Simiformes, and ATXN3L3 in Cercopithecidae, in addition to ATXN3L (herein called ATXN3L1) originated in Haplorrhini. ATXN3 and ATXN3L1 are both under purifying selection throughout primate evolution, maintaining about 70% of amino acid identity. Also, the high conservation of ATXN3L1 Josephin domain hints at functional redundancy with the parental disease-associated ATXN3. ATXN3L2 presents a remarkable nucleotide similarity to ATXN3 (79%) in an interrupted reading frame, which may produce a regulatory RNA. Conversely, ATXN3L0 is likely a non-functional retrocopy and ATXN3L3 is absent in humans with no relevance for the disease. The comparison of (CAG)_n interruption patterns of the different paralogs in several primates elucidates the process leading to the currently observed pure long tracts in human ATXN3, responsible for disease when expanded. This study intends to pioneer the identification of new paralogs of SCA-associated genes and the use of phylogenetic analyses to explore their potential role for targeted therapies.

Significance.

Polyglutamine spinocerebellar ataxias are human progressive brain diseases caused by the abnormal expansion of repetitive DNA sequences in particular genes. These expansions lead to the production of faulty proteins that accumulate in nerve cells, ultimately causing their dysfunction and degeneration over time. Previously, other researchers have found that similar counterparts of these disease-associated genes in the human genome (gene copies) are able to compensate for the loss of normal protein function, offering potential therapeutic targets.

In this study, we discovered new copies of the gene responsible for the most frequent of these dominant ataxias worldwide. Our findings highlight an approach that can be followed to identify similar gene copies linked to other repeat-based disorders and select the most promising therapeutic candidates.

Introduction

Gene duplication contributes to the evolution of species by the acquisition of new genetic material for mutation, genetic drift, and selection to act upon (Innan and Kondrashov 2010). Traditionally, duplicate genes are reported to originate through a DNA-based mechanism, often via unequal crossing over (from segmental or tandem duplications), thus, maintaining a similar organization and physical proximity to the parental gene (Kaessmann 2010). However, RNA-based duplication, termed retrotransposition or retroduplication, represents another source of gene duplication, thereby generating intronless copies (retrocopies) that can be located elsewhere in the genome (Kaessmann et al. 2009). In this instance, the processed mRNA of the parental gene undergoes reverse transcription and random incorporation by the long interspersed nuclear elements-1 (LINE1 or L1) machinery. Given that retrocopies generally lack transcriptional elements, they were thought to be non-functional copies designated as retropseudogenes or processed pseudogenes. Therefore, they were long believed to be free from selective pressures, accumulating diverse loss of function mutations such as premature stop codons and frameshift mutations (Kaessmann et al. 2009). Notwithstanding, there are functional retrogenes with retained or acquired promoter elements, which potentiate their transcription (Casola and Betrán 2017).

The identification of gene paralogs (originated through DNA- or RNA-based duplications) is not only crucial to obtain a complete picture of the evolution of organisms but also to identify potential pathological buffering mechanisms that might be applied as therapies when the parental gene is associated to disease (Felício et al. 2023). Spinocerebellar ataxias (SCAs) are a heterogeneous group of neurodegenerative disorders characterized by progressive cerebellar ataxia, resulting in lack of muscle control and coordination. In this group of autosomal dominant disorders, seven are classified as polyglutamine ataxias since they result from the elongation of a polyglutamine (polyQ) tract encoded by an expanded CAG repeat (Klockgether et al. 2019). Those are SCA1, SCA2, MJD (Machado–Joseph disease)/SCA3, SCA6, SCA7, SCA17, and dentatorubral-pallidoluysian atrophy (DRPLA). More recently, researchers found a new CAG repeat expansion in the coding region of THAP11 associated with spinocerebellar ataxia in two Chinese families, which may suggest that the group of polyQ SCAs will expand in the future (Tan et al. 2023). In these disorders, elongated polyQ stretches promote protein misfolding and aggregation, which inhibit the native function and/or lead to abnormal protein interactions (Klockgether et al. 2019).

Interestingly, studies in SCA1 demonstrated that a highly conserved DNA-based copy of the human disease-associated ATXN1 gene (called ataxin-1 like; ATXN1L; Mizutani et al. 2005) was able to partially compensate the loss of wild-type ATXN1 function, modulating the cytotoxicity of expanded ATXN1 and suppressing SCA1 neuropathology in mice (Bowman et al. 2007; Crespo-Barreto et al. 2010; Carrell et al. 2022). Indeed, ATXN1 and ATXN1L share a conserved AXH (ataxin-1 and HMG-box protein 1) domain (66% identity), essential for their interaction with each other and with the transcriptional corepressor Capicua (Mizutani et al. 2005; Kim et al. 2013). In short, the overexpression of ATXN1L enabled the paralog to compete with the mutant ATXN1 for the interaction with Capicua, partially rescuing the transcriptional complex, a known mechanism that drives cerebellar pathology in SCA1 (Lam et al. 2006; Bowman et al. 2007).

The most common dominant ataxia worldwide is MJD/SCA3 (Sequeiros et al. 2012; Martins and Sequeiros 2018). Similar to other polyQ disorders, its causing gene, ataxin-3 (ATXN3, 14q32.12) contains a CAG repeat, which normally ranges from 12 to 44 CAGs. Conversely, affected individuals usually present expanded alleles surpassing the 61 repeats, with the expansion size partially explaining the variability in the age-at-onset, which can vary from infancy to later in adulthood (average onset at 40 years old) (Conceição and Manuela 2011). Thus, modifiers of disease phenotype must play an important role in MJD. The native function of wild-type ATXN3 has been associated with proteostasis, through its deubiquitinase activity (Burnett et al. 2003; Burnett and Pittman 2005), and with transcriptional regulation activities by its binding to DNA and several transcription-related factors (Li et al. 2002; Evert et al. 2006).

ATXN3 belongs to an old gene lineage found in non-bilateria organisms, suggesting that this gene performs essential functions (Sousa e Silva et al. 2023). A human paralog of ATXN3 is already known to exist in primates (ATXN3L; Xp22.2) (Scheel et al. 2003; Vlasschaert et al. 2017; Sousa e Silva et al. 2023), but very few studies have investigated ATXN3L from an evolutionary standpoint; thus, ATXN3L still lacks a full taxonomic annotation, the same way that other paralogs may remain unknown. No ATXN3L protein has been detected so far, but an in vitro experiment demonstrated that, if translated, ATXN3L may present higher deubiquitinase activity than the parental ATXN3. Interestingly, the replacement of three ATXN3 residues (S12F, R59L, and T60A) was found to almost reach the level of ATXN3L activity (Weeks et al. 2011). Since both proteins show distinct ubiquitin recognition sites (Weeks et al. 2011), it would be interesting to explore evolutionary constraints of ATXN3/ATXN3L residues and their biological relevance.

To this end, we aimed to reconstruct the evolutionary history of ATXN3 family/paralogs and to conceptually explore their roles as MJD modifiers. To this end, we conducted an exhaustive analysis of ATXN3 homologs by retrieving the sequences of 33 primate species and other close mammals from genomic databases. Our analyses allowed us to identify four main retrocopies of ATXN3 in primates, originated at different time points in mammalian evolution. To explore the relevance of ATXN3 copies, we examined sequence diversity, gene structure, and selective pressures acting on ATXN3 paralogs during evolution. Our results suggest that ATXN3L (herein called ATXN3L1) and the newly identified ATXN3L2 are the most promising paralogs to proceed with functional characterization and explore their role and interaction with the parental gene. Finally, our in-silico approach aims to pioneer the study of other SCA-associated paralogs and their potential to affect and/or partially maintain the function of the respective parental genes.

Results

ATXN3 Retrotransposition Events in Primates

Upon the analysis of the sequence similarity between the coding sequence of the human ATXN3 (hsaATXN3) and 33 primate genomes, we identified four common retrocopies named ATXN3L0 to ATXN3L3 (Fig. 1, Table S1) through the analysis of synteny (genes flanking each retrocopy; Fig. 1a) and sequence identity between orthologs and paralogs (Table S2a). Next, we estimated the timing of each retrotransposition based on their most recent common ancestor in the primate or mammalian species tree (Fig. 1): ATXN3L0 (81.3 to 91.0 MYA), ATXN3L1 (formerly known as ATXN3L; 61.6 to 71.1 MYA), ATXN3L2 (40.0 to 44.2 MYA), and ATXN3L3 (15.50 to 19.80 MYA), next described in detail.

Fig. 1.

Reconstruction of ATXN3 gene family evolution in primates. a) Local synteny information of the different ATXN3 paralogs in primate species, with the top rows showing synteny observed in each retrocopy for most species followed by exceptions. Interrogation points (?) represent lack of annotation in NCBI assemblies. Slashes (/) represent complete loss of synteny in the respective direction. b) Primate phylogenetic tree showing the origin of the different ATXN3 copies (obtained from TimeTree, https://timetree.org/). The presence of ATXN3 paralogs is marked in each species with colored circles. A square bracket ([) indicates two copies from a common ancestor. The stars represent the time at which the retrotranspositions seem to have occurred. MYA, million years ago.

ATXN3L0 is the only copy present in most primates, with possible gene loss events occurred in chimpanzees (Pan troglodytes and Pan paniscus) and tarsiers (Carlito syrichta; Table 1, Fig. 1b). Several premature stop codons observed along the ATXN3L0 phylogeny indicated that this is likely an ancient processed pseudogene (located 42 Mb away from the parental gene in humans; LOC100418768; 14q21.3; Table S1). Among non-primate mammals, we identified sequences likely to be homologous to ATXN3L0, namely in the closest evolutionary relatives to primates (flying lemurs and treeshrews) and among rodents (in the Sciuridae family).

Table 1.

Four main ATXN3 paralogs identified in primate species (NCBI database)

Species name	ATXN3L0	ATXN3L1	ATXN3L2	ATXN3L3
Homo sapiens	LOC100418768	ATXN3L	LOC100132280	─
Pan paniscus	─	ATXN3L	LOC100968165	─
Pan troglodytes	─	LOC129138637	LOC100613414	─
Gorilla gorilla gorilla	NA	ATXN3L	LOC101124755	─
Pongo abelii	NA	ATXN3L	LOC100455664	─
Pongo pygmaeus	NA	ATXN3L	LOC129042124	─
Hylobates moloch	NA	ATXN3L	LOC116813850	─
Nomascus leucogenys	NA	ATXN3L	LOC100601102	─
Macaca mulatta	NA	LOC711119	LOC700434	LOC699321
		LOC114674839a
Macaca fascicularis	NA	ATXN3L	LOC102117545	LOC102119590
Macaca thibetana thibetana	NA	ATXN3L	LOC126960647	LOC126930683
		LOC126945336a
Macaca nemestrina	NA	ATXN3L	LOC105476127	LOC105494606
		LOC105478274a
Mandrillus leucophaeus	NA	LOC105550045	LOC105555234	LOC105527940
		LOC105550026a
Cercocebus atys	NA	ATXN3L	LOC105584571	LOC105590961/LOC105590962
Papio anubis	NA	ATXN3L	LOC110743999	LOC101017058
		LOC101021868a
Theropithecus gelada	NA	ATXN3L	LOC112630196	LOC112635516
Chlorocebus sabaeus	NA	LOC103231613	LOC103237716	LOC119623360/LOC103239276
		LOC119619931a
Colobus angolensis palliatus	NA	ATXN3L	LOC105506288	LOC105516342/LOC105516344
Piliocolobus tephrosceles	NA	ATXN3L	LOC111549645	LOC111537833
Rhinopithecus bieti	NA	LOC108523901/LOC108544961	LOC108541821	LOC108529641
		LOC108523899a	…	…
Rhinopithecus roxellana	NA	ATXN3L	LOC104664224	LOC115900089
		LOC104669352a	…	…
Trachypithecus francoisi	NA	ATXN3L	LOC117079069	LOC117089275
Cebus imitator	NA	ATXN3L	LOC108291766	─
Sapajus apella	NA	ATXN3L	LOC116529997	─
Saimiri boliviensis boliviensis	NA	ATXN3L	LOC101048375	─
Aotus nancymaae	NA	ATXN3L	LOC105710782	─
Callithrix jacchus	NA	ATXN3L	─	─
Carlito syrichta	─	LOC103268037	─	─
Microcebus murinus	NA	─	─	─
Propithecus coquereli	LOC105813271	─	─	─
Lemur catta	NA	─	─	─
Otolemur garnettii	NA	─	─	─
Nycticebus coucang	NA	─	─	─

In some species, different contigs (separated by a slash) were found to represent the same gene; NA, not annotated; ─, not found.

^aTandem duplicates of ATXN3L1.

ATXN3L1 (Xp22.2 in humans) displayed 69% to 80% sequence identity among all Simiiformes where it was identified (Fig. 1b, Table 1 and Table S2a; exceptions in Fig. S1). Tarsiers were considered to have an ATXN3L1 copy (despite lacking synteny evidence, Fig. 1a) given that Carlito syrichta sequence showed 72% identity to hsaATXN3 (similar to values observed for the remaining orthologs; Fig. 1, Table S2a). This highlights the importance of integrative analyses since a different synteny from the consensus does not invalidate gene paralogy (as also shown by the annotated ATXN3L from Theropithecus gelada; Fig. 1a). Notably, in Carlito syrichta, we found a second copy of ATXN3L1 (LOC103249688), sharing 95% sequence similarity, compatible with a recent species-specific retrotransposition event (Table S2a). Finally, other ATXN3L1 tandem duplication events seem to have occurred in a Cercopithecidae ancestor (Fig. 1b). Duplicates are located about 17 to 22 Kb from ATXN3L1 and conserve intact reading frames, most of them displaying a very high sequence similarity (99% to 100%) with the respective ATXN3L1 source gene. While analyzing local synteny, we also noted the lack of consensus in current annotation regarding the location of ATXN3L1 and its tandem duplicate between species (Table S1).

We identified ATXN3L2 (LOC100132280; 8q23.2 in humans) in most Simiiformes (Table 1, Fig. 1b) exhibiting an average nucleotide identity of 79% with the respective parental ATXN3 (Table S2a). Nevertheless, in contrast to ATXN3L1, ATXN3L2 presents premature stop codons in all analyzed species.

ATXN3L3 was identified only in Cercopithecidae, displaying high similarity with the respective ATXN3 paralog (87% to 97% nucleotide identity; Fig. 1b, Table S2a), but different open reading frames are predicted among Old-World monkeys (primates that belong to the family Cercopithecidae). This younger retrotransposed gene is absent in Homoinidea species.

Finally, we detected additional events of ATXN3 retrotransposition in different primates (Theropithecus gelada, Chlorocebus sabaeus, Aotus nancymaae, Callithrix jacchus, and Carlito syrichta) and in the common ancestor of Nycticebus coucang and Otolemur garnettii (Lorisoidea; 34.5 to 40.7 MYA), all classified by us as independent retrocopies (Fig. 1b, Table S1).

(CAG)_n Region of ATXN3 Paralogs

To address possible mechanisms leading to human-specific (CAG)_n expansion in ATXN3 we have analyzed the repetitive tract across the identified ATXN3 orthologs and paralogs. Briefly, we aligned the repeat regions of ATXN3, ATXN3L1, ATXN3L2, and ATXN3L3 for all primate sequences available, taking into consideration that all may present some level of polymorphism. We excluded ATXN3L0 from this analysis because no (CAG)_n tract was detected in this retrocopy.

The ATXN3 repeat, which in humans is characterized by the (CAG)₂ CAA AAG CAG CAA (CAG)_n motif, is apparently conserved in half of the analyzed primate species, with only five species showing pure glutamine tracts (CAG/CAA codons) with variable lengths (Table 2).

Table 2.

Configuration of the (CAG)_n tract in primate genome reference sequences of ATXN3 and its retrocopies

Species	ATXN3	ATXN3L1	ATXN3L2	ATXN3L3
Homo sapiens	(CAG)2 CAA AAG CAG CAA (CAG)8	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	(CAG)3 (CGG CAG)8	─
Pan paniscus	(CAG)2 CAA AAG CAG CAA (CAG)14	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	CAG (CGG CAG)7 GGG CAG	─
Pan troglodytes	(CAG)2 CAA AAG CAG CAA (CAG)12	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	CAG (CGG CAG)8 GGG CAG	─
Gorilla gorilla gorilla	(CAG)2 CAA AAG (CAG)4	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	(CAG)10 GAG	─
Pongo abelii	(CAG)2 CAA AAG CAG CAA (CAG)2 CCG CAA (CAG)2 CAA (CAG)2 CCG CAA (CAG)7	CAT (CAG)2 GAA CAG AAG CTG CAG (CAA)2 CAG	(CAG)6	─
Pongo pygmaeus	(CAG)2 CAA AAG CAG CAA (CAG)2 CCG CAA (CAG)2 CAA (CAG)2 CCG CAA (CAG)7	CAT (CAG)2 GAA CAG AAG CTG CAG (CAA)2 CAG	(CAG)6	─
Hylobates moloch	(CAG)2 CAA AAG CAG CAA (CAG)7	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	(CAG)5	─
Nomascus leucogenys	(CAG)3 CAA (CAG)8	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	(CAG)8	─
Macaca mulatta b	(CAG)2 CAA (CAG)2 AAG (CAG)7	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	(CAG)6 CAC CAG	(CAG)2 CAA AAG CAG CAA (CAG)2 AAG (CAG)5 GAG (CAG)2
Macaca fascicularis	(CAG)2 CAA (CAG)2 AAG (CAG)7	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	(CAG)6 CAC CAG	(CAG)2 CAA AAG CAG CAA (CAG)2 AAG (CAG)5 GAG (CAG)2
Macaca thibetana thibetana b	(CAG)2 CAA (CAG)2 AAG (CAG)11	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	(CAG)6 CAC CAG	(CAG)2 CAA AAG CAG CAA (CAG)2 AAG (CAG)5 GAG (CAG)2
Macaca nemestrina b	(CAG)2 CAA (CAG)2 AAG (CAG)10	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	CAG CAA (CAG)4 CAC CAG	(CAG)2 CAA AAG (CAA)2 (CAG)2 AAG (CAG)5 GAG (CAG)2
Mandrillus leucophaeus b	(CAG)2 CAA AAG CAG CAA (CAG)2 AAG (CAG)10	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	(CAG)6 CAC CAG	(CAG)2 CAA AAG (CAG)5 GAG (CAG)2
Cercocebus atys	(CAG)2 CAA AAG CAG CAA (CAG)2 AAG (CAG)7	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	CAG CAT (CAG)4 CAC CAG	(CAG)2 CAA AAG (CAG)5 GAG (CAG)2
Papio anubis b	(CAG)2 CAA AAG CAG CAA (CAG)2 AAG (CAG)7	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	(CAG)7 CAC CAG	(CAG)2 CAA AAG (CAG)5 GAG (CAG)2
Theropithecus gelada a	(CAG)2 CAA AAG CAG CAA (CAG)2 AAG (CAG)7	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	CAG CAT (CAG)5 CAC CAG	(CAG)2 CAA AAG (CAG)5 GAG (CAG)2
Chlorocebus sabaeus a b	(CAG)2 CAA AAG CAG CAA (CAG)8 AAG (CAG)3 AAG (CAG)6	CAT CGG CAG GAA CAG AAG (CAG)2 (CAA)2 CAG	(CAG)5 (CAC)3 CAG	(CAG)2 CAA AAG CAG CAA (CAG)2 AAG (CAG)5 GAG (CAG)2
Colobus angolensis palliatus	Absent	CAT (CAG)2 GAA CAG AAG (CAG)4 CAA CAG	(CAG)6 CAC CAG	(CAG)2 CAA AAG CAG CAA (CAG)6 GAG (CAG)2
Piliocolobus tephrosceles	(CAG)2 CAA AAG (CAG)4 AAG (CAG)5	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	(CAG)3 CAT (CAG)2 CAC CAG	(CAG)2 CAA AAG CAG CAA (CAG)2 AAG (CAG)8 GAG (CAG)2
Rhinopithecus bieti b	(CAG)2 CAA AAG CAG CAA (CAG)2 AAG (CAG)5	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	(CAG)8 CAC CAG	GAG CAG CAA AAG CAG CAA (CAG)2 AAG (CAG)3 AAG (CAG)2 CAA (CAG)3 GAG (CAG)2
Rhinopithecus roxellana b	(CAG)2 CAA AAG CAG CAA (CAG)2 AAG (CAG)5	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	(CAG)9 CAC CAG	GAG CAG CAA AAG CAG CAA (CAG)2 AAG (CAG)3 AAG (CAG)6 GAG (CAG)2
Trachypithecus francoisi	(CAG)2 CAA AAG CAG CAA (CAG)2 AAG (CAG)5	CAT (CAG)2 GAA CAG AAG (CAG)2 (CAA)2 CAG	(CAG)7 CAC CAG	GAG CAG CAA AAG CAG CAA (CAG)5 GAG (CAG)7
Cebus imitator	CAG CAA (CAG)4 (CAA)2 CAG CAA (CAG)8	GAT GAG CAA GAA CAA (CAG)2 AGG AAA CAA AAG	GAG (CAG)3 CAA	─
Sapajus apella	CAG CAA (CAG)4 (CAA)2 (CAG)8	GAT GAG CAA GAA CAA (CAG)2 AGG AAA CAA AAG	GAG (CAG)3 CAA	─
Saimiri boliviensis boliviensis	CAG CAA (CAG)5 (CAA)5 CAG CAA (CAG)12	GAC GAG CAA GAA CAA (CAG)2 AGG AAA CAA AAG	GAG (CAG)4 CAA	─
Aotus nancymaae a	Absent	CAT GAG CAA GAA CAA (CAG)2 AGG AAA CAA AAG	GAG (CAG)3 CAA	─
Callithrix jacchus a	CAG CAA (CAG)5 CAA (CAG)3 GAG (CAG)3 GAG (CAG)3	CAT GAG CAA GAA CAA (CAG)2 AGG AAA CAA AAG	─	─
Carlito syrichta a	(CAA)2 (CAG)2	CAG TGG CGG CAG (CAC)2 CAG GAC (CAC)4 TGC CAA CAG	─	─
Microcebus murinus	(CAG)17 AAG	─	─	─
Propithecus coquereli	Absent	─	─	─
Lemur catta	(CAG)8 CGG (CAG)3	─	─	─
Otolemur garnettii a	(CAG)5 CAA (CAG)5 AAG CGG GAG	─	─	─
Nycticebus coucang a	CAG CTG (CAG)3 AAG (CAG)5 AAG CAG CGG GAG	─	─	─

─, not found.

^a ATXN3 recent species-specific retrocopies: Theropithecus gelada LOC112629965 - (CAG)₂ CAA AAG CAG CAA (CAG)₂ AAG (CAG)₈, Chlorocebus sabaeus LOC119623504: CAT CGG CAG GAA CAG AAG (CAG)₂ (CAA)₂ CAG, Aotus nancymaae LOC105706409 - CAG CAA (CAG)₁₂, Callithrix jacchus LOC100388307 - CAG CAA (CAG)₃, Callithrix jacchus LOC100393745 - (CAG)₇ CAA (CAG)₅, Carlito syrichta LOC103249688 - CAG CAA (CAG)₂, Otolemur garnettii unknown copy: absent, Nycticebus coucang LOC128584333 - CAG CAA (CAG)₂ CAA.

^bTandem duplicates of ATXN3L1: Macaca nemestrina LOC105478274 - AAG CTA -AT ATT (CAG)₂ GAA CAA (AGC)₃ (AAC)₂ -AG, Chlorocebus sabaeus LOC103231613 - TTG CTC GGC AGG GAA CAG AAG (CAG)₂ CAA CAG CGC -AG. Macaca mulatta LOC711119, Macaca thibetana thibetana LOC126945336, Mandrillus leucophaeus LOC105550026, Papio anubis LOC101021868, Rhinopithecus bieti LOC108523899 and Rhinopithecus roxellana LOC104669352 tandem duplicates have the same (CAG)_n configuration as their counterpart ATXN3L1 represented in the table.

ATXN3L1 displays a shorter and more interrupted repeat in comparison to ATXN3 (Table 2). Notably, we found the ATXN3L1 repeat motif to be conserved in most Catarrhini (CAT (CAG)₂ GAA CAG AAG (CAG)₂ (CAA)₂ CAG), whereas New-World monkeys (from the Platyrrhini parvorder) and Carlito syrichta presented more complex repetitive regions.

ATXN3L2 presents a polymorphic hexanucleotide (CGG CAG)_n flanked by CAG and GGG CAG triplets in humans and chimpanzees (Table 2), while the remaining species have mostly pure or simpler CAG repeats.

The ATXN3L3 repetitive region is very similar to the one in ATXN3, presenting the same initial pattern (CAG)₂ CAA AAG CAG/CAA except for Asian colobine species (Rhinopithecus and Trachypithecus; Table 2), which display more interrupted sequences.

Characterization of Retrocopies in the Human Genome

In humans, whereas ATXN3L0 and ATXN3L2 are in different intergenic regions (chr14 and chr8, respectively; Fig. 2a and 2c), ATXN3L1 shares genomic coordinates with an antisense long non-coding RNA (lncRNA) gene in chromosome X (Fig. 2b).

Fig. 2.

Genomic characterization of ATXN3 human paralogs. UCSC Genome Browser view displaying summarized regulatory and chromatin-related features (ENCODE, ORegAnno), transcription start site (TSS; FANTOM5), and poly(A) sites (custom track from PolyASite v.2.0) for a) ATXN3L0, b) ATXN3L1, and c) ATXN3L2.

As anticipated, ATXN3L0 (ENSG00000258639) consists of two exons without a putative transcription start site (TSS) and a polyadenylation (polyA) signal, which indicates it is unlikely to be transcribed (Fig. 2a).

ATXN3L1 (ENSG00000123594) is a single exon gene that overlaps the last three exons of GS1-600G8.3 (ENSG00000231216.1), an unknown antisense gene of 17 exons that produces a lncRNA (Fig. 2b). According to the GTEx portal, GS1-600G8-3 (here named ATXN3L1-AS1) is mainly expressed in testis, mirroring the expression pattern observed for the ATXN3L1 transcript. Consistent with this, TSS activity mapped to the beginning of ATXN3L1 5′UTR (chrX:13,320,050-13,320,053; TTAT) was detected exclusively in adult testis. Both ReMap Atlas and ORegAnno databases indicate that this 5′UTR region is enriched for binding sites of multiple transcription factors. Additionally, a polyadenylation site within the ATXN3L1 3′UTR (chrX:13,318,643-13,318,669; ATTAAA@-21) was also used exclusively in testis, supporting tissue-specific transcriptional regulation (Fig. 2b).

ATXN3L2 (ENSG00000253107) is also reported as a single exon with two predicted TSS (hg_90280.1 chr8:110555387-110555397; AGACAAATAAA; and hg_90281.1 chr8:110556281-110556292; AAGCAGCAGCAG) (Fig. 2c). Indeed, near the 5′ and 3′UTRs there are possible regulatory regions reported in ENCODE, ORegAnno and ReMap. However, there are no ESTs aligned to ATXN3L2, no polyA signal close to its 3′UTR, and no gene expression is reported in GTEx portal (Fig. 2c).

Protein Conservation and Selective Tests

The predicted human ATXN3L1 protein sequence presents 70% of global identity with ATXN3 (Table S2b), preserves its catalytic sites (C14, H119, N134) and nuclear export signs (Fig. 3b) as well as the Josephin Domain (JD) at the N-terminal and three ubiquitin-interacting motifs (UIMs), at the C-terminal region (Tables S2c and S2d). Among haplorrhines, UIMs 1 to 3 show 83, 63% and 79% identity, respectively; however, unlike UIMs 1 and 2, the UIM3 is not always present, with both ATXN3 and ATXN3L1 missing this motif in a few primates due to alterations in the coding sequence (eg Aotus nancymaae ATXN3 and Trachypithecus francoisi ATXN3L1; Fig. 3a; Fig. S1). Interestingly, in Macaca thibethana thibetana there seems to have occurred the insertion of an intron (Fig. 3a; Fig. S1) similar to one found in some ATXN3 alternative splicing transcripts identified in humans (Bettencourt et al. 2010). On the other hand, the nuclear localization signal (NLS) sequence seems to be absent in ATXN3L1, raising the question of the subcellular location of this paralog.

Fig. 3.

Conservation of ATXN3L1. a) Comparison of ATXN3 and ATXN3L1 conserved protein motifs (predicted in MEME suite) among primate species. b) Protein alignment of Homo sapiens ATXN3 and ATXN3L1 sequences, highlighting important domains, ATXN3 catalytic sites (Q9, C14, H119, N134—red stars), and three mutations (S12F, R59L, and T60A; called triple mutation—blue circles) found to almost match the ATXN3L1 and ATXN3 deubiquitinase activity. JD, Josephin domain. UIM, ubiquitin-interacting motif. SUIM-assoc, unstructured region C-term to UIM in ATXN3. NES, nuclear export signal. NLS, nuclear localization signal. PolyQ, polyglutamine stretch.

To assess the intensity of the selective forces acting on both ATXN3 and ATXN3L1, we performed several phylogenetic-based tests using codeml (PAML package), after removing the (CAG)_n region from sequence alignments. First, to evaluate the evolution of both genes in primates, we estimated a single ω ratio for the entire phylogeny (one-ratio model), in which we assumed no differentiation in selective constraints of ATXN3 and ATXN3L1 (ω_total = 0.23, Fig. 4a). Then, to examine whether the parental gene and retrocopy were subjected to different selective pressures, we applied a two-ratio model, which considers two branches within the phylogeny—the ATXN3 or ATXN3L1 clades. The parental gene is under stronger selective constraints (ω_ATXN3 = 0.22) when compared to ATXN3L1 (ω _ATXN3L1 = 0.41, P = 7.77E−03; Fig. 4a). To investigate the strength of the selective constraints in ATXN3L1, we employed the RELAX model from the HyPhy software, which infers if relaxed or intensified selection has been acting along the selected branches by introducing a selection intensity parameter (k). Here, we validated that purifying selection has become less stringent along ATXN3L1 branches in comparison with ATXN3 (k = 0.00, LR = 60.79, P = 0.00E + 00).

Fig. 4.

ATXN3L1 shows an overall more permissive negative selection than the ATXN3 gene clade. a) Model comparison of omega ratios calculated for the ATXN3-ATXN3L1 tree—one-ratio (M0) versus two-ratio and free-ratio models. b) Cladogram of ATXN3 and ATXN3L1 gene lineages demonstrating PAML-estimated ω values under the free-ratio branch model, which assumes independent values on each branch. Values of infinity (∞) indicate branches lacking synonymous substitutions, and values of 0.0001 indicate branches lacking non-synonymous substitutions. LnL, log-likelihood. np, number of parameters. df, degrees of freedom. LRT, Likelihood ratio test.

Finally, we applied the free-ratio branch model to estimate a single omega value per each branch of the tree combining the two genes (Fig. 4b), which showed significant variation in ω values across the different branches (ATXN3-ATXN3L1, −2ΔLnL = 195.91; likelihood ratio test [LRT] P-values = 0.00; Fig. 4a). Overall, our results suggest strong selective constraints (negative selection) acting on ATXN3 (ω = 0.044) prior to the event of retrotransposition, later followed by a period of a slight weakening in selective pressures driving the evolution of ATXN3 (ω = 0.3534) and ATXN3L1 (ω = 0.3917, Fig. 4b). Next, the selective forces were restored in ATXN3 to the original strength in Tarsiiformes (ω = 0.066) and Simiformes (ω = 0.0396) ancestors, with several branches lacking non-synonymous substitutions (ω = 0.0001, Fig. 4b). For ATXN3L1, the selection was maintained as less restrictive but still compatible with functional conservation in both Platyrrhini (ω = 0.3935) and Catarrhini (ω = 0.9959) ancestors, with the existence of a period of accelerated evolution for the Rhinopithecus ancestor (ω = 1.6137, Fig. 4b).

Next, to identify regions in ATXN3 and ATXN3L1 that could have undergone adaptive evolution in primates, we performed site-models tests for each gene separately. According to our results, only the parental gene shows some evidence of diversifying evolution in some codons (Table 3: M1a vs. M2a; M7 vs. M8; Fig. 5), with less than 1% of ATXN3 sites showing high omega values (M8 vs. M7 p1 = 2.95E−03, ω = 297.94). Indeed, only 11 codons have a posterior probability (pp) > 0.5 for the positive-selection class (Table 3). When considering merely the sites with a pp > 70%, just five codons have a high probability of being positively selected: 292, 312, 313, 349, and 350. Three of them are located between UIM2 and UIM3 domains, in a disordered region flanking the polyQ repeat (Fig. 3b and 5). Codon 292 encodes the first residue downstream the polyQ, and codons 312 and 313 are close to the start of ATXN3 UIM3. Codons 349 and 350 encode additional C-terminal residues specific to orangutan (Pongo pygmaeus, Pongo abelii), gibbons (Hylobates moloch, Nomascus leucogenys), and tarsier (Carlito syrichta) (Fig. 5). In fact, these species have alterations at the end of exon 11, resulting in the extension of the coding sequence (Fig. S1), which makes these two C-terminal amino acids unlikely to have a great functional impact.

Table 3.

Parameter estimates and likelihood scores under different branch and site models

Tree	Model	Ln L	np	Model comparison	−2ΔLnL	df	Estimates of parameters				LRT P-value	Positive sites
ATXN3 (n = 26)	M3	−2919.22	55	M0 versus M3	79.24	4	P:	0.89	0.11	0.00	0.00E+00	…
							ω:	0.03	0.78	309.75		…
	M0	−2958.84	51				ω0:	0.10				…
	M2a	−2919.50	54	M1a versus M2a	21.35	2	P:	0.91	0.09	0.00	2.31E−05	…
							ω:	0.04	1.00	317.32		…
	M1a	−2930.17	52				P:	0.91	0.09	…		…
							ω:	0.04	1.00	…		…
	M8	−2922.21	54	M7 versusM8	25.17	2	p0 = 1.00	P = 0.01	q = 0.03	…	3.42E−06	240 I 0.560, 268 M 0.623, 292 G 0.921, 297 Q 0.570, 306 A 0.649, 312 L 0.815, 313 G 0.720, 320 M 0.503, 338 R 0.553, 349–0.985, 350–0.785
							(p1 = 2.95E−03)	ω= 297.94	…	…
	M7	−2934.79	52				P=	0.02	q=	0.11		…
	M8a	−2930.12	53	M8a versusM8	15.83	1	p0 = 0.92	P = 0.57	q = 10.79		6.93E−05	…
							(p1 = 0.08)	ω= 1.00				…
ATXN3L1 (n = 24)	M3	−3735.68	51	M0 versus M3	36.65	4	P:	0.57	0.41	0.01	2.13E−07	…
							ω:	0.13	0.71	3.43		…
	M0	−3754.01	47				ω0:	0.37				…
	M2a	−3736.13	50	M1a versus M2a	2.23	2	P:	0.73	0.26	0.01	3.28E−01	…
							ω:	0.19	1.00	3.85		…
	M1a	−3737.24	48				P:	0.72	0.28	…		…
							ω:	0.18	1.00	…		…
	M8	−3735.82	50	M7 versusM8	5.23	2	p0 = 0.98	P = 0.64	q = 1.07	…	7.31E−02	…
							(p1 = 0.02)	ω= 3.32	…	…		…
	M7	−3738.43	48				P=	0.475	q=	0.71		…
	M8a	−3737.24	49	M8a versusM8	2.85	1	p0 = 0.72	P = 21.48	q = 99.00		9.14E−02	…
							(p1 = 0.27)	ω= 1.00				…

LnL, log-likelihood; np, number of parameters; df, degree of freedom; LRT, Likelihood ratio test.

Fig. 5.

Positively selected sites in ATXN3. ATXN3 sites under positive selection identified by M8 site-model from PAML (pp > 0.5), and Fixed Effect Likelihood model (FEL; P ≤ 0.1) and the Mixed Effects Model of Evolution (MEME; P ≤ 0.05) from HyPhy/Datamonkey. Highlighted residues are colored according to PAML posterior probability scores (light gray: pp > 0.5, dark gray: pp > 0.7, black: pp > 0.9). Codons predicted by the two models are enclosed in a box. Propithecus coquereli, Lemur catta, Colobus angolensis palliatus, Cebus imitator, Mandrillus leucophaeus, Aotus nancymaae and Rhinopithecus bieti were not included in positive selection analyses due to sequences constraints.

For ATXN3L1, no codons were identified as experiencing positive selection (M1a vs. M2a: 2ΔLnL = 2.23, P = 3.28E−01; M7 vs. M8: 2ΔLnL = 5.23, P = 7.31E−02; M8 vs. M8a: 2ΔLnL = 2.85, P = 9.14E−02).

To corroborate PAML results for ATXN3, we additionally used two models implemented through HyPhy/Datamonkey: Fixed Effect Likelihood model (FEL, with P ≤ 0.1) and Mixed Effects Model of Evolution (MEME, with P ≤ 0.05). The first model assumes a constant selective pressure for each site, while the second presume that selective pressures between branches are uncorrelated or independent. FEL model identified 93 codons under purifying selection distributed along the ATXN3 protein, mostly in the N-terminal that includes the Josephin domain and nuclear export signs (from codon 1 to 180; Table S3a). Interestingly, we found codon 313 to be undergoing diversifying selection throughout the phylogeny (β = 1.652, LRT = 2.843, P = 0.09) (Table S3a), as predicted by PAML (pp = 0.720). In parallel, the MEME model showed codon 306 (LRT = 6.80, P = 0.01) as an episodic positive selected site (Table S3b), also predicted in PAML site model (pp = 0.649). When accounting for the results from both PAML and HyPhy, we observed that the positively selected sites are in the C-terminal region of ATXN3, which includes UIM3 (Fig. 5).

Discussion

In this study, we identified four main retrotransposition events of ATXN3. Besides the previously reported ATXN3L1, we identified an unannotated pseudogene probably originated in Euarchontoglires (here named as ATXN3L0) and two not fully characterized genes (herein called ATXN3L2 and ATXN3L3) that arose in Simiformes and Cercopithecidae ancestors, respectively (Table 1, Fig. 1b). Additional retrotransposition events occurred more recently, either in a single species or in a restricted clade (Table S1). These findings are not surprising given several studies suggesting a high rate of retrotransposition in mammals, especially in primates, probably driven by their enrichment in full-length LINE 1 elements (Ivancevic et al. 2016). Studies pointed to periods of intense activity of transposons around the mammalian radiation, in Simiiformes (40.0 to 44.2 MYA), Haplorrhini (61.6 to 71.1 MYA), and Eutheria (placental mammals, 94.7 to 101.9 MYA) ancestors (Marques et al. 2005; Pace and Feschotte 2007), in concordance with age estimates of ATXN3 retrotransposition. Also, during the search for ATXN3L0 copies in Mammalia species, no retrocopies were found in Prototheria, in concordance with the previously report that genomes from Monotremata species have no LINE1 elements (Ivancevic et al. 2016). On the other side, in disagreement with the lack of evidence for the activity of transposons after the emergence of Cercopithecidae (with no LINE1 younger than 37 million years; Pace and Feschotte 2007), we identified recent retrotransposition events, which may indicate that a few active elements were able to be transmitted within these primates.

The identification of sequences likely to be homologous to ATXN3L0 in some non-primate mammals though absent among rodents (with the exception of the Sciuridae family) could mean that (i) a large-scale loss of ATXN3L0 occurred in rodents related species, (ii) rapid evolution of ATXN3L0 made this retrocopy no longer identifiable in some species from Supraprimates, or (iii) some independent retrotransposition events of ATXN3 occurred in some species resulting in similar homologs when compared with ATXN3L0.

ATXN3L1 was first thought to have originated in a Catarrhini ancestor (Weeks et al. 2011), and later in the Simiiformes infraorder, after the separation of the Tarsiiformes lineage (Sousa e Silva et al. 2023); however, in this study, we established the time of origin of ATXN3L1 to a Haplorrhini ancestor (Fig. 1b), approximately 25.7 million years earlier. The absence of ATXN3L1 tandem copies in some Old-World monkeys, as well as ATXN3L0 and ATXN3L2 retrocopies in a few species is possibly due to gene loss events after speciation (Table 1 and Fig. 1b), although the possibility of incomplete genomic data cannot be excluded. Besides, we should also consider that gene losses may be a species-specific event, not occurred in other species from the same order or family. Another difficulty we faced on the identification of paralogs/orthologs were discrepancies in synteny, which may be expected once chromosomal rearrangements are known to have occurred after speciation (Chiou et al. 2022), as seen in Theropithecus gelada, as we found synteny blocks from chr. X differently positioned in comparison to Papio anubis (Fig. 1a and Table S1).

Interestingly, ATXN3L1 was retrotransposed from an autosome to the X chromosome, contrarily to most reports of X-linked genes being transposed to autosomal chromosomes, with mammalian X chromosomes producing a disproportionately high number of functional retroposed genes in comparison to autosomes (Emerson et al. 2004; Kaessmann 2010). Still, ATXN3L1 seems to exhibit the same testis-biased expression as previously described for X-linked retrotransposed genes. Previous studies have discussed the role of testis in the process of gene evolution, allowing young paralogs to evolve, fixate, and later broaden their expression pattern in somatic tissues with time (Kaessmann 2010; Guschanski et al. 2017). Additionally, studies have been reporting an intrinsic relation between testis and brain expression (Guo et al. 2003), where these tissues share numerous common proteins, evidencing mutual biochemical and functional features (Matos et al. 2021). This way, we cannot discard the hypothesis of ATXN3L1 being expressed in other tissues given the lack of functional studies involving this retrogene, including a precise quantitative transcript expression analysis. For instance, if this retrogene is expressed in specific structures or cell types of the adult brain, or even during stages of brain development (not previously observed), this paralog could gain further relevance in the study of Machado–Joseph disease. It is certain, however, that ATXN3L1 is a transcriptionally active gene, which either retained the ATXN3 promoter during the retrotransposition event (Okamura and Nakai 2008) or acquired a nearby promoter element (the 5′ closest gene to ATXN3L1 is EGFL6; enriched in placenta, according to GTEx). In humans, ATXN3L1 retrogene is located within an antisense long non-coding RNAs (lncRNAs) missing annotation and functional characterization (Fig. 2b). It is known that, in the human genome, some retrocopies overlap with lncRNAs, regulatory RNA sequences of >200 nucleotides that can establish specific interactions with RNA molecules and proteins involved in the expression of the parental gene (Bryzghalov et al. 2016). Therefore, it would be crucial to characterize the targets of this unknown lncRNA gene and its effect on RNA expression, especially on sequences highly similar to ATXN3.

Additionally, the fact that ATXN3L1 is located on the X chromosome may be interesting to explore gene expression dosage and how this could potentially be relevant to MJD pathogenesis. Mammals have evolved a compensatory mechanism to randomly inactivate one of the female X chromosomes in somatic cells, however, a number of genes located at the p-arm escape X inactivation (15% to 25% of all known ChrX genes; Balaton et al. 2015). These genes are called XCI escape genes or escapees and, in humans, the vast majority is located on the short arm-Xp, as ATXN3L1 (Xp22.2) (Disteche 1999). Although no differences in MJD phenotype are known between male and female patients (Coutinho 1992), it would be interesting to explore this possibility.

Both ATXN3 and ATXN3L1 seem to be under purifying selection, with the retrocopy experiencing less selective constraints than the parental gene (Fig. 4a). This is not surprising since most genes involved in severe monogenic diseases are reported to be under strong negative selection (Blekhman et al. 2008). Besides, duplicates of genes with high biological relevance have often been observed under more relaxing constraints, thus promoting functional divergence (eg subfunctionalization) (Jordan et al. 2004); still, a retained purifying selection allows the maintenance of functional utility, in accordance with our results for ATXN3L1 (Fig. 4b). Interestingly, previous studies have shown that the orthologs evolutionary rates for duplicates have been negatively correlated with the number of paralogs and the strength of selection between paralogs (Jordan et al. 2004); this may indicate that the number of ATXN3 paralogs is leading to an overall conservation of all copies.

The hypothesis of ATXN3L1 being conserved and functionally relevant is reinforced by its high similarity with ATXN3: the Josephin domain is putatively maintained, as well as the two essential ubiquitin-interacting motifs (UIM1 and UIM2, the only ones present in the shorter isoform of human ATXN3), and the potential nuclear export signal (Fig. 3). Although there are no studies on the subcellular localization or molecular targets of ATXN3L1, the conservation of the catalytic triad (C14, H119, N134) and nuclear export signs when compared with ATXN3 is a good indication of its deubiquitinase function in the cytoplasm. Indeed, the ATXN3L1 Josephin domain adopts the same overall fold as the ataxin-3 Josephin domain, and in vitro assays demonstrated its catalytic activity (Weeks et al. 2011), which is not surprising given the 85% sequence identity between the two domains. At the same time, we cannot discard the hypothesis of new functions being ascribed to ATXN3L1, which would be of upmost importance to help clarifying the biological functions of wild-type (non-expanded) ATXN3, still unknown.

Intriguingly, despite the similarities between ATXN3 and ATXN3L1, only the parental protein seems to contain some fast-evolving residues. Those are all located in the C-terminal region of ATXN3, which, according with the three-dimensional model includes a disordered region, followed by the polyQ and the UIM3 (Fig. 5). These ATXN3 residues can be under diversifying selective pressures that alter protein properties related to the ubiquitin interaction, or the region around UIM3 may not be critical for its activity and thus not subjected to negative constraints. Indeed, shorter ATXN3 protein-coding transcripts that include only UIM1 and UIM2 have been shown to be more abundant in the brain, while UIM3-containing transcripts were more frequent in blood (Raposo et al. 2024). Therefore, this pattern of selective pressures in ATXN3 may be linked to Machado–Joseph disease pathogenesis.

ATXN3L2 presents a high nucleotide sequence conservation when compared with ATXN3 (Table S2a), but multiple frameshift mutations and premature stop codons during primate evolution may have turned it into a non-functional processed pseudogene. This does not prevent, however, the transcription of ATXN3L2 (Fig. 2c), which may play a role in expression regulation of the parental gene. In fact, in SCA7, it was previously proposed that the retropseudogene ATXN7L3B (which accumulated frameshift mutations since its duplication, resulting in premature stop codons and a truncated open reading frames [ORF]) acts as a post-transcriptional regulator of ATXN7, contributing to specific neurodegeneration (Tan et al. 2014). Also, previous studies have demonstrated that processed pseudogenes are capable of regulating the expression of parental genes through different mechanisms such as (i) antisense transcripts forming stable RNA duplexes with the parental mRNAs inhibiting the translation, (ii) long noncoding RNAs (lncRNAs) acting as molecular sponges for RNA-binding proteins (RBPs) or microRNAs through shared binding sites, and (iii) small-interfering RNA (siRNAs) acting as a decoy for transcription factors (Cheetham et al. 2020).

Finally, the analysis of ATXN3 repeat sequences in other non-human primates and paralogs provided some insight into the mutational mechanisms that underlies the expansion of the CAG repeat associated with MJD in humans. The observation that only certain repeat sequences become unstable was suggested to be related to sequence-specific DNA structures, potentially crucial for instability. In ATXN3, gene regions surrounding the (CAG)_n were shown to be highly conserved in primates, whereas the mutation rate at the two codons adjacent to the repeat was estimated to be 27 times higher than in intronic regions (Djian et al. 1996). The fact that ATXN3L1 exhibits more interrupted and stable (CAG)_n configurations over primate evolution suggests that the ancestral (CAG)_n region of ATXN3 was highly interrupted. However, mutations in ATXN3 over the last 61.6 to 71.1 MYA likely led to a less interrupted (CAG)_n region, increasing its susceptibility for expansion. In addition to the analysis of configuration patterns, it would be interesting to study the polymorphic nature of repetitive regions in all retrocopies in several primates.

In this study, we reconstructed the history of ATXN3 gene family and performed a comprehensive evolutionary analysis to gain insight into the functional relevance of ATXN3 paralogs. ATXN3 was described three decades ago, and it has been extensively explored due to its implication in Machado–Joseph disease (MJD/SCA3), the most frequent dominant ataxia worldwide. Interestingly, increased retrotransposition activity has been suggested to be linked to neurodegenerative disorders (Terry and Devine 2020). This may be related to the fact that brain is the only known somatic tissue where retrotransposons are de-repressed in humans, whereas, in other tissues, silencing mechanisms restrict retrotransposition due to its inherited capacity to disrupt genomes and transcriptomes (Yang and Wang 2016). Several studies indicate that retrotransposed genes may be important players in pathogenesis by influencing transcriptional and post-transcriptional expression of disease-related genes (Ciomborowska-basheer et al. 2021). Our study highlights the value of evolutionary analyses to gain insight into the potential functional relevance of unexplored duplicates. This approach enables a more effectively identification of the most promising candidates for subsequent in vitro assays and experimental testing of their roles as disease modifiers.

Materials and Methods

Identification of ATXN3 Homologous Genes and Orthology/Paralogy Inference in Primate Species

We conducted tblastn searches against 33 representative and RefSeq annotated primate assemblies (Table S4; Supplemental files S1 and S2) in the Basic Local Alignment Search Tool (BLAST, https://blast.ncbi.nlm.nih.gov/Blast.cgi; accessed July 28, 2023) (Camacho et al. 2009) from National Center for Biotechnology Information (NCBI) using Homo sapiens reference ATXN3 protein sequence (NP_004984.2) as input. To identify genes that might be missing from previous studies and databases, we inferred orthology and paralogy through local synteny analysis in Genome Data Viewer (https://www.ncbi.nlm.nih.gov/genome/gdv/) from NCBI.

Search for ATXN3L0 Orthologs in Mammalian Species

To ascertain the origin of ATXN3L0 retrotransposition event, we performed tblastn searches on other 58 representative genomes of mammals (accessed October 18, 2023; Table S4) using the predicted protein sequence obtained for Homo sapiens (LOC100418768). The focus was on species from the Euarchontoglires superorder.

Collection of Nucleotide and Protein Sequences

The sequences of the genes identified in the tblastn analysis, together with their annotation data, were directly retrieved from NCBI's GenBank nucleotide database. In addition, homologous sequences from Gallus gallus and Anas platyrhynchos were used as outgroup in the for the complete nucleotide alignment with all paralogs.

Paralogs and orthologs sequences were aligned using MUSCLE implemented in Geneious Prime software (2022.2.2, www.geneious.com) and manually curated/inspected according with sequence conservation among species.

We used Augustus gene prediction algorithm implemented in Geneious Prime to predict ORF and potential coding regions. We obtained protein sequences from the translation of collected nucleotide sequences and calculated nucleotide and protein identity percentage (%) matrices in Geneious Prime.

Phylogenetic Analysis

To determine the best nucleotide substitution model for tree estimation, we used jModelTest version 2.1.10 (Darriba et al. 2012). For ATXN3, ATXN3L1, and ATXN3-ATXN3L1 (phylip files, without CAG regions), we identified the TIM1 + G (transition model with proportion of invariable site and rate heterogeneity across sites; code 012230), GTR + G (general time reversible model with rate heterogeneity across sites), and GTR + I + G (general time reversible model with proportion of invariable sites and rate heterogeneity across sites) models, respectively, as the most suitable ones. When considering the alignment with all nucleotide sequences and the (CAG)_n region, the TVM + I + G (transversion model with proportion of invariable sites and rate heterogeneity across sites; code 012314) model was found to be the best-fitting model.

We inferred the maximum likelihood tree using the PhyML algorithm implemented in Geneious Prime under default parameters (analyses were performed with 1,000 bootstrap searches); and next displayed the phylogenetic trees using FigTree v1.4.4 (http://tree.bio.ed.ac.uk/software/figtree/).

Selective Pressure Analysis in ATXN3 and ATXN3L1

To evaluate the selective pressures acting on ATXN3 and ATXN3L1, we employed the codeml program from Phylogenetic Analysis by Maximum Likelihood (PAML) (Yang 2007) package using the EasyCodeML tool (Gao and Ho 2019). Sequences were only included after the removal of positions affected by premature stop codons and frameshift mutations. To test for variable selective pressures among branches, we performed the branch model using either the null model (one ratio) or the nested model (two-ratio) (Bielawski and Yang 2003). We calculated the ratio of nonsynonymous (dN) to synonymous (dS) substitution rates (ω, omega), assuming distinct neutral or selective models, with ω = 1, >1 or <1 expected in a neutral, positive, or purifying (negative) evolution, respectively. Models implemented in codeml vary in terms of their assumptions about ω variation across the sequence or across branches of the phylogeny. Firstly, we used branch models (one-ratio vs. free-ratio, one-ratio vs. two-ratio) to test the hypothesis of equal evolutionary rates between ATXN3 (background) and ATXN3L1 (foreground) gene lineages after retrotransposition. The Free-ratio model allowed to analyze selection constraints in all species for both ATXN3 and ATXN3L1 genes. Next, we used site-specific (M0 vs. M3, M1a vs. M2a, M7 vs. M8, M8 vs. M8a) models to test if certain codons of ATXN3 and ATXN3L1 were under positive selection in haplorrhines. For branch models each newick (.nwk) tree was altered to label the clade of interest as foreground (inserted with $) and the remaining branches as background.

We used LRTs to perform pairwise comparisons between codon models and obtained significant results from twice the variation of likelihoods (−2Δl) using χ² tests. If LRT yielded a significant result for M7 versus M8, the Bayes empirical Bayes procedure was used to identify codons/amino acids under positive selection (posterior probability ≥ 0.5).

Additionally, we used the Hypothesis Testing using Phylogenies (Hyphy) software implemented in the Datamonkey Adaptive Evolution Server (https://www.datamonkey.org/) (Weaver et al. 2018) to detect sites under selection in codon-based alignments with the Fixed Effect Likelihood model (FEL) and the Mixed Effects Model of Evolution (MEME) (Kosakovsky Pond and Frost 2005; Anisimova 2012; Murrell et al. 2012). The FEL model is based on a maximum-likelihood approach to infer dN and dS substitution rates, assuming that the selection pressure for each site is constant along the phylogeny. The MEME model employs a mixed-effects maximum likelihood approach where the ω on each branch is independent of those from other branches, ie selective pressures between branches are uncorrelated. This way, each site has two ω rate classes and corresponding weights representing the probability of the site to evolve under ω rate class, at a given branch. To infer ω rates, MEME infers a single α (dS) value and two separate β (dN) values, β− and β+, which share the same α, per site. In short, MEME model allows ω to vary between sites and branches, detecting both episodic and pervasive positive selection. We also used LRTs to ascertain significant results (P ≤ 0.1 and P ≤ 0.05, respectively). Next, test whether the strength of natural selection has been relaxed or intensified along the ATXN3L1 clade in comparison with ATXN3, we used the RELAX model (Wertheim et al. 2015). After fitting a codon model with three ω parameters (ω0 ≤ω1 ≤ 1 ≤ω2) to the entire phylogeny (null model), the RELAX model determined the proportion of sites in ATXN3L1 (foreground) and ATXN3 (background) branches using a branch-site model. Additionally, a selection intensity parameter (k) was introduced to compare a null model (k = 1) with an alternative model, assessing if the strength of natural selection has been intensified (k > 1) or relaxed (k < 1) in the ATXN3L1 branch relative to the ATXN3 branch. Statistical confidence (P ≤ 0.05) was calculated with LRTs and the Holm-Bonferroni correction.

Identification of ATXN3L1 Conserved Motifs

To identify conserved motifs possibly essential for gene function, we used the Multiple EM (Expectation Maximization) for Motif Elicitation (MEME) tool (https://meme-suite.org/meme/tools/meme, version 5.5.4) from MEME Suite (Bailey and Elkan 1994; Bailey et al. 2009). Parameters adjusted for MEME analysis included: number of repetitions (zero or one occurrence per sequence (zoops)), maximum number of motifs (15, ordered by E-value) and width of sequence (20 to 200 amino acids). We used the ATXN3-ATXN3L1 protein fasta file aligned and concatenated as input to analyze conserved regions likely to be contributing to protein structure and function.

Genomic Characterization of ATXN3 Human Paralogs

We performed the genomic characterization of ATXN3 human paralogs (ATXN3L0, ATXN3L1, and ATXN3L2) through the UCSC Genome Browser (https://genome.ucsc.edu/; Nassar et al. 2023) displaying summarized regulatory and chromatin-related features from ENCODE project (cis-regulatory elements, DNase I hypersensitivity peaks, and H3K4Me1/H3K4Me3/H3K27Ac histone marks), and ORegAnno database (predicted regulatory elements). Also, we assessed predicted promoters (EPDnew v006), transcription start sites (TSS; FANTOM5), and poly(A) sites (custom track from PolyASite v.2.0; https://www.polyasite.unibas.ch/; Herrmann et al. 2020). Along these transcriptional indicators, we used human mRNAs from GenBank to verify the previous identified RNAs of these paralogs.

Supplementary Material

Supplementary material is available at Genome Biology and Evolution online.

Data Availability

All datasets were derived from sources in the public domain. The data and analysis underlying this article are available in the article and in its online Supplementary material.